ClawHub
Back to skill

Security audit

agentic-ai-research

Security checks across malware telemetry and agentic risk

Overview

This skill performs web-based Agentic AI literature research and only writes local wiki files after explicit confirmation, with no evidence of hidden exfiltration or destructive behavior.

Install if you want a Chinese-oriented Agentic AI literature review workflow. Before enabling wiki compilation, check the absolute paths shown in the confirmation step and keep them inside the current project, not your home directory or a sensitive folder.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Tp4

High
Category
MCP Tool Poisoning
Confidence
87% confidence
Finding
The skill markets itself primarily as a web-based literature review tool, but it also references a bundled `wiki-creator` component that can initialize directories, scan local wiki content, parse local files, and generate indexes and manifests. Even though the wiki step is described as optional and gated by confirmation, this extends the skill into broader local file processing and modification, which increases the attack surface and could surprise users who only expected web search and markdown output.

Intent-Code Divergence

Medium
Confidence
90% confidence
Finding
The docstring promises the tool will never write to a global home directory, but the code does not enforce that invariant when a user supplies --root. An attacker or mistaken caller could pass a path under the home directory or another sensitive location, causing unintended file creation and violating the documented safety boundary.

Vague Triggers

Medium
Confidence
77% confidence
Finding
The trigger conditions are broad enough to activate on ordinary research-related requests such as asking for papers or a survey, which can cause the skill to run unexpectedly. In context, that matters because the skill has file read/write and network permissions, so accidental invocation could lead to unanticipated data access, network activity, or creation of local wiki artifacts.

Natural-Language Policy Violations

Medium
Confidence
92% confidence
Finding
The template hard-codes Chinese as the primary output language without any indication that the user's preferred language should be honored. In an agent skill, this can override user intent and cause unauthorized transformation of outputs into a different language, reducing usability and potentially changing nuance in research summaries or citations.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.